.jpg){kind=link}
The answer: conflicting leaders do not establish an overnight reversal
UDN’s October28 Taipei release puts Puma Shen at45% and Chiang Wan-an at41%; ETtoday’s release the previous evening puts Chiang at48.6% and Shen at41.1%. Different leaders matter, but consecutive publication dates do not establish a one-day shift. Field periods overlap and methods differ.[1][6]
Neither choosing favorable figures nor declaring the Ocean Current finished answers the evidence. Senge asks what could reflect measurement, what could reflect real preferences and how far public information allows us to distinguish them. This is an assessment of evidence, not a recalculation of election results.
Start with field dates and coverage
UDN says it interviewed1,115 adult Taipei residents on October20–22 and25 through random-digit sampling from residential and mobile telephone frames, weighting by sex, age and district. UDN financed the survey.[6]
TVBS and Mirror Daily describe ETtoday’s October21–25 closed online questionnaire:1,462 valid responses from eligible adult Taipei members of its own database, invited by text message and email, financed by Eastern Online Polling. Receiving an invitation on a phone does not make it a random mobile-telephone interview.[1][2]
Telephone coverage and nonresponse need scrutiny; a member survey also requires scrutiny of membership, geographic labels and participation. Either mode may be informative, but differences in disclosure do not quantify the cause of disagreement or demonstrate fabrication.
What can be compared?
| Item | UDN October28 release [6] | ETtoday October27 release, as reported [1][2] | Limit |
|---|---|---|---|
| Field dates | October20–22,25 | October21–25 | Overlapping periods, not a one-day panel |
| Frame | Residential and mobile telephone | Closed member-database online questionnaire | Different coverage and response mechanisms |
| Valid responses | 1,115 | 1,462 | A larger sample does not eliminate selection bias |
| Preference | Shen45%, Chiang41% | Chiang48.6%, Shen41.1% | Design-specific estimates, not counted ballots |
| Published margin | ±3.0 points | ±2.56 points | Neither total error nor a universal test of the lead |
Why a simple average or exact vote-flow claim is inadequate
A midpoint can look like compromise while hiding decisions about weights, field periods and comparability. Without those choices, a simple average is not a validated forecast. This article supplies neither an unsupported uncertainty interval nor a claim that sampling error captures all error.
A single-proportion margin also cannot automatically test a candidate difference. Relationships between estimates, design effects and weights matter. People outside a membership frame do not become adequately represented simply because more members answer. Apply these questions equally to favorable and unfavorable findings.
Cross-sectional recall of past votes is not longitudinal tracking of actual choices. Multiplying a recall percentage by a historical total would not establish an exact number of switched votes; recall, response and participation are separate selection processes.
Separate preference, expectations and candidate responses
Newtalk and Nextapple describe UDN’s different directions for preference and expected winner. Voters can prefer one person and expect another to win. Undecided status, campaign interest and satisfaction likewise cannot substitute for turnout intention or winning probability.[3][4]
In a separate TVBS interview, Shen emphasizes actual voting and local contact. That is a candidate’s response, not independent evidence that his strategy works. His comments on refusal composition do not replace a full nonresponse analysis. Fair commentary retains both the political voice and analytical distance.[5]
Agreement, disagreement and alternatives
Five reports help compare wording and methods, not supply five independent measurements. TVBS and Mirror Daily share the ETtoday wave; Newtalk, Nextapple and the TVBS interview share UDN’s wave. UDN supplies a direct release for one survey. Consistent reporting reduces transcription concerns without removing underlying bias.
Nextapple quotes a commentator interpreting several surveys as favorable movement for Shen. That interpretation does not establish a house effect or a spiral of silence. Alternatives include actual movement, coverage, refusal, willingness to express preferences, question presentation and weighting. They may coexist; available evidence cannot allocate precise shares.[4]
More discriminating evidence would include repeated comparable designs, disclosed question order and weighting, and scrutiny of policy costs, responsibilities and timetables. These are further research questions, not causal effects already measured by Senge.
Conclusion: support attention with inspectable choices
Different leaders increase uncertainty and the need for transparent comparison. Shen’s momentum still faces participation and implementation tests; Chiang’s lead in another survey likewise cannot replace answers to voters. Analysis should not oscillate between a completed miracle and total collapse for attention.
Useful commentary returns readers to evidence: who was contacted, how and when they were asked, and who remains unobserved. Answer those questions before discussing the November28 choice, so surveys do more than confirm existing positions.